Root-zone soil moisture estimation under forest canopy using SAR and optical fusion
C-band radar stops at the canopy surface; L-band SAR from ALOS-2 PALSAR-2 and the forthcoming NISAR mission penetrates to rooting depths where tree water stress actually begins. Fused with Sentinel-2 vegetation indices and a water-cloud model, it yields volumetric soil moisture estimates that no optical sensor can match.
Sensors
- ALOS-2 PALSAR-2 (JAXA): L-band (1.27 GHz) SAR with 3–10 m single-look resolution in spotlight and stripmap modes; revisit approximately 14 days. L-band wavelength (~23 cm) penetrates closed forest canopies and is sensitive to dielectric changes in the top 5–15 cm of mineral soil, reaching the upper root zone. Dual-polarisation (HH/HV) separates volume from surface scattering contributions.
- NISAR (NASA/ISRO, planned launch 2025): Dual-frequency L-band and S-band SAR; 12-day repeat, global coverage. L-band channel designed explicitly for ecosystem and soil moisture science. Expected to provide consistent L-band time series comparable to PALSAR-2 but with denser temporal sampling, improving seasonal moisture-change detection under canopy.
- Sentinel-1 C-SAR (ESA): C-band (5.405 GHz), 10 m IW mode, 6-day revisit over most land. Penetration depth in moist mineral soil is typically 2–5 cm, making it a surface moisture indicator rather than a root-zone sensor. Used in fusion schemes to characterise canopy water content and constrain the vegetation scattering term in water-cloud models.
- Sentinel-2 MSI (ESA): 10–20 m multispectral, 5-day revisit. Red-edge bands (705 nm, 740 nm) and near-infrared provide NDVI, EVI and the red-edge chlorophyll index used to parameterise canopy water content and leaf area index, both required inputs for separating vegetation from soil backscatter in SAR retrieval models.
- SMAP (NASA): L-band radiometer at 36 km spatial resolution, 2–3 day revisit. Too coarse for stand-level work but provides a calibration anchor for landscape-scale soil moisture, and its published retrieval algorithms (the tau-omega model) inform the parameterisation approaches applied at finer resolution using PALSAR-2.
Why C-band stops at the litter layer
Radar backscatter is governed by the dielectric constant of the target, and the dielectric constant of soil rises steeply with liquid water content. The problem for forest managers is getting the signal to the soil in the first place. C-band wavelengths (roughly 5 cm) are scattered heavily by canopy elements: leaves, small branches, the wet litter mat. Penetration depth in a closed broadleaf canopy is limited to the upper canopy volume. What returns to the satellite is mostly vegetation scatter, with soil information buried beneath it.
L-band wavelengths (roughly 23 cm) interact less strongly with foliage and more strongly with large woody stems and the soil surface beneath them. Under moderately dense forest, a meaningful fraction of the L-band signal reaches mineral soil and returns through the canopy, carrying dielectric information from depths of 5 to 15 cm in mineral soils, depending on moisture state and soil texture. That range covers the upper root zone for most tree species. It is not a deep-soil sensor, but it is the only spaceborne radar that routinely reaches below the litter layer in closed-canopy conditions.
The water-cloud model and why it needs Sentinel-2
The water-cloud model (Attema and Ulaby, 1978) treats the canopy as a uniform cloud of water droplets. Total backscatter is the sum of direct vegetation scatter, two-way attenuation through the canopy, and the attenuated soil contribution. The model has two vegetation parameters, A and B, which must be estimated from independent data. Leaf area index and canopy water content, both retrievable from Sentinel-2 red-edge and near-infrared bands, serve as proxies for these parameters. Without the optical constraint, the model is under-determined: you cannot separate how much of the L-band signal came from the canopy versus the soil.
In practice, a retrieval scheme ingests a time series of PALSAR-2 HH and HV backscatter alongside coincident or near-coincident Sentinel-2 NDVI or the red-edge chlorophyll index. The vegetation parameters are fitted over a calibration period when soil moisture is independently known (from rain-gauge networks or SMAP), then held fixed while soil moisture is retrieved from the residual soil backscatter term. Published studies in boreal and tropical forests report root-mean-square errors of roughly 0.04 to 0.08 m³/m³ volumetric moisture content under favourable conditions. That range widens considerably under dense litter or high soil organic matter.
Where the retrieval breaks down
Organic-rich soils present a fundamental ambiguity. Peat and thick forest litter have dielectric properties that change with moisture differently from mineral soils. The empirical relationships between dielectric constant and volumetric water content that underpin retrieval algorithms (Topp's equation being the most widely used) were calibrated on mineral soils. Applying them to organic horizons introduces systematic bias that is difficult to correct without site-specific calibration data. Retrieval errors in peatland-edge forests can exceed 0.10 m³/m³.
Dense litter layers also attenuate L-band differently from standing canopy biomass, and this attenuation is not captured by Sentinel-2 vegetation indices, which see only the green canopy above. Steep terrain introduces geometric distortion in SAR imagery (foreshortening, layover) that corrupts backscatter values on slopes greater than roughly 20 degrees. Finally, the 14-day PALSAR-2 revisit means that fast-moving moisture pulses after convective rainfall may be missed entirely between acquisitions. NISAR's denser repeat cycle should help, but the satellite was not operational at the time of writing.
What a moisture map actually tells a forest manager
A root-zone moisture layer updated at each PALSAR-2 pass gives forest managers a spatially explicit picture of where trees are approaching water stress before foliar symptoms appear in optical imagery. Drought stress in conifers, for example, typically becomes visible in Sentinel-2 shortwave-infrared bands only after several weeks of deficit. The SAR-derived moisture signal precedes that by a full phenological lag, giving time to prioritise irrigation in plantation blocks or to adjust prescribed burn windows based on actual fuel moisture state rather than meteorological proxies.
For national forest inventories and carbon accounting programmes, the moisture layer feeds into forest productivity models and helps explain inter-annual variation in growth increments that would otherwise be attributed incorrectly to disturbance. It is also an input to drought-vulnerability assessments, though the page covering tropical forest moisture stress and drought vulnerability mapping addresses that application in detail.
Turning physics into a delivered product
A working retrieval pipeline has several moving parts: PALSAR-2 scene download and radiometric terrain correction, Sentinel-2 atmospheric correction and index computation, temporal co-registration, water-cloud model inversion, and uncertainty propagation through to the output moisture grid. Each step has failure modes. Radiometric terrain correction requires a digital elevation model of adequate resolution; the freely available Copernicus DEM at 30 m is generally sufficient for gentle terrain. Cloud cover over the optical acquisition is irrelevant to the SAR data but leaves gaps in the vegetation parameterisation, which must be filled by temporal interpolation or climatological averages.
Satellize runs this fusion pipeline on open-constellation data and can add commercial PALSAR-2 tasking on client licence for priority areas. The Tonga crop-estimation programme demonstrated the team's approach to fusing SAR and optical time series in a data-sparse environment, and the same co-registration and model-inversion workflow transfers directly to forest soil moisture retrieval. Outputs are delivered as georeferenced GeoTIFF moisture grids with per-pixel uncertainty estimates, compatible with standard GIS platforms.
Typical figures
| Spatial resolution (L-band SAR input) | 3–10 m (PALSAR-2 spotlight/stripmap); expected 25 m nominal for NISAR science products |
| Spatial resolution (output moisture grid) | Typically 25–100 m after speckle filtering and model inversion; finer outputs carry higher uncertainty |
| Revisit (PALSAR-2) | 14 days; off-nadir tasking can reduce this to 2–4 days for priority sites at additional cost |
| Revisit (Sentinel-1 C-band, supporting) | 6 days at equator; 1–3 days at high latitudes |
| Radar frequency (primary) | L-band: 1.27 GHz (PALSAR-2); ~23 cm wavelength |
| Soil penetration depth | 5–15 cm in mineral soils (L-band); 2–5 cm (C-band); depth decreases with increasing moisture |
| Retrieval accuracy (published range, mineral soils) | RMSE ~0.04–0.08 m³/m³ volumetric water content under favourable conditions; >0.10 m³/m³ on organic soils |
| PALSAR-2 archive depth | From 2014 (ALOS-2 launch); predecessor ALOS PALSAR from 2006 |
| Cloud sensitivity | SAR acquisitions unaffected by cloud; Sentinel-2 optical inputs require cloud-free compositing |
| Delivery format | GeoTIFF moisture grids with per-pixel uncertainty band; optional WMS/WMTS tile service |
Analytics Satellize can run
| Root-zone volumetric soil moisture map | Water-cloud model inversion on PALSAR-2 HH/HV backscatter, constrained by Sentinel-2 LAI and canopy water content | GeoTIFF raster layer per acquisition date, with uncertainty band |
| Seasonal moisture anomaly layer | Z-score normalisation of moisture time series against multi-year PALSAR-2 climatology | Monthly anomaly GIS layer flagging pixels more than 1.5 standard deviations below seasonal mean |
| Canopy water content index (Sentinel-2) | Shortwave-infrared ratio and red-edge index retrieval from atmospherically corrected Sentinel-2 reflectance | 10–20 m raster, used as vegetation parameterisation input and delivered as standalone product |
| Water-cloud model parameter map (A, B coefficients) | Calibration against SMAP landscape-scale anchor and available in-situ rain-gauge data during wet-season period | Static coefficient rasters for client study area, updated annually or after major disturbance |
| Drought stress precursor alert | Threshold exceedance on moisture anomaly layer, cross-checked against 14-day rainfall deficit from ERA5 reanalysis | Automated alert report (PDF and GeoJSON) identifying stands at risk, issued within 48 hours of PALSAR-2 processing |
| Terrain-corrected SAR backscatter archive | Radiometric terrain correction using Copernicus 30 m DEM; gamma-naught normalisation | Analysis-ready backscatter time series (GeoTIFF stack) from 2014 to present for client AOI |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.